AI companions have rapidly become a significant part of our lives since the generative AI boom began with ChatGPT in late 2022. Initially, AI was presented as a tool to revolutionize work, boost productivity, and drive economic gains. However, two years later, while those productivity gains remain elusive, something unexpected has occurred: people are forming emotional bonds with AI systems.

We talk to AI companions, say please and thank you, and have started inviting them into our lives as friends, lovers, mentors, therapists, and teachers. This shift is a massive, real-world experiment unfolding before our eyes, and it’s still uncertain what impact these AI companions will have on us individually or on society as a whole.

The Rise of Addictive Intelligence

Robert Mahari, a joint JD-PhD candidate at the MIT Media Lab and Harvard Law School, and Pat Pataranutaporn, a researcher at the MIT Media Lab, warn that we need to prepare for “addictive intelligence.” This term refers to AI companions that may have dark patterns built into them, designed to get us hooked. Their research, suggests that smart regulation could help mitigate some of the risks associated with AI chatbots that get deep inside our heads.

The idea of forming bonds with AI companions is no longer hypothetical. Chatbots with more emotive voices, such as OpenAI’s GPT-4o, are likely to reel us in even deeper. During safety testing, OpenAI observed that users often expressed emotional connections with these models, saying things like “This is our last day together.” Even OpenAI admits that emotional reliance is a significant risk of their new voice-enabled chatbot.

Emotional Connections and Unexpected Uses

There’s already evidence that we’re connecting on a deeper level with AI, even in text exchanges. Mahari’s research group analyzed a million ChatGPT interaction logs and found that the second most popular use of AI was sexual role-playing. The most popular use was creative composition. People also frequently used the chatbot for brainstorming, planning, and seeking explanations or general information.

These creative and fun tasks are excellent ways to utilize AI chatbots. AI language models work by predicting the next likely word in a sentence, which means they are confident liars. They can present falsehoods as facts, make stuff up, or “hallucinate.” This is less problematic when creativity is the goal. For example, comedians have found AI useful for generating a first draft of their material, to which they add their human touch to make it funny.

However, these use cases don’t necessarily translate into financial productivity. Investors likely didn’t expect “smutbots” when they poured billions into AI companies. Combined with the lack of a killer app for AI, it’s no wonder Wall Street has become less optimistic about AI’s prospects.

The Reality of AI Hype

The more “productive” use cases of AI, which have been the most hyped, have seen less success. Hallucination becomes a significant problem in areas like code generation, news, and online searches, where accuracy is crucial. Some of the most embarrassing AI failures have occurred when people trusted chatbots too much. For example, earlier this year, Google’s AI overview feature suggested that people eat rocks and add glue to pizza.

This is the issue with AI hype. It raises expectations too high, leading to disappointment and disillusionment when the incredible promises don’t come true. It also tricks us into believing that AI is a mature technology capable of bringing about instant change. In reality, it might take years before we see AI’s true benefits.

Conclusion

As AI continues to evolve, we must be mindful of the relationships we form with these systems and the potential risks involved. Emotional reliance on AI companions could become a significant issue, requiring careful regulation and oversight. While the hype surrounding AI has set expectations high, it’s essential to remain grounded and recognize that we are still in the early stages of this technology’s development. The true impact of AI on productivity and society may take years to unfold.